Nowadays automatic interaction systems are developed and new possibilities for communication appeared. Digitalization of various information becomes a common pattern, and with the development in technology simple sentences or commands can be processed directly in the machine systems. Automatic chat-bots, question-answer systems and various computer solutions that support man-machine interactions need e_cient methods for classification if the input is correct. The correctness of the input secures that it shall be formulated by a human and not by another machine which is just trying to break into the system. In this article we present a model for language processing to verify some basic aspects of correctness. For the proposed processing an input is decomposed by applied calculation of language descriptors, which are forwarded to the probabilistic neural network processor for validation. The proposed model is examined and presented in examples for English language, and the numerical results are discussed in terms of possible advantages and disadvantages in security aspects.